The All-OBP Team Revisited
When Don gets curious about something, neuron activity increases in a lot of minds and computer processors begin to overheat. So what is Don curious about now? How OBP and SLG interacts.
More data to fuel the fire ...
A couple of months back I lost what was left of my membership card in the "sabermetric
cabal" by suggesting that a team whose OBP and SLG were each .425 wouldn’t
score a gazillion runs.
The actual number, as some of you will remember, was 956. That projection was
arrived at by the use of a basic version of Bill James’ contribution
to our understanding of the mechanics of run scoring (Runs Created).
Several voices chimed in to correct this estimate, including BBBA co-authors
Mike Emeighand Sean Forman. While others simply trotted out their
own pet formulae for projecting run scoring, Mike did some real work, using
recent (1999-2000) actual game data to demonstrate that there is indeed more
of a synergistic effect for "high OBP/low SLG" teams than vice-versa.
Mike is still tinkering with this some more, so I won’t try to formalize his
results just yet, but what I can tell you is that teams with "high OBP/low
SLG" tend to outperform their RC estimates. Teams with the opposite profile
("low OBP/high SLG") tend to underperform.
The variation isn’t quite as high as those who proposed that a .425/.425 team
would produce close to 1200 runs in a 162-game season, however. And even with
Mike’s excellent work, I wasn’t satisfied to stop there.
So I called on the masterful data manipulator of latter-day sabermetrics, Tom
Ruane. I asked Tom to go through his 22-year database of play-by-play data,
and create an actuarial chart of run scoringsorted by OBP and SLG range
combinations. In addition to his exceptional talents, Tom is one of the truly
good guys in the field, and so he graciously obliged me with yet another of
my massively-scaled requests for help.
The OBP and SLG range combinations traced the run scoring in actual games from
1979-2000 (over 130,000 team game records). They were constructed as follows:
after a category of .000-.199, SLG was broken out into 10-point increments (.200-.209,
.210-.219) on up to .690-.699. Everything over .700 was lumped into one final
category. For OBP, the initial category was .000-.099, followed by 10-point
increments beginning with .100-.109 and continuing in that fashion up to .490-.499.
All games where OBP was over .499 were lumped into one final category.
If I’ve managed to explain this with any level of clarity, you should see where
we’re headed. The idea was to look at all of the OBP/SLG combinations, and total
up the number of games and the number of runs scored in each. Dividing the latter
(runs) by the former (games) would give us the average number of runs/game in
each OBP/SLG combination.
For the purposes of the discussion of an "all-OBP" team like the
one proposed by BBBA reader Duane Thomas, which featured a series of
.400+ OBP walkmen (and whose combined statistics in the single seasons chosen
worked out to .290 BA/.425 OBP/ .425 SLG), we are interested in only one area
on Tom’s vast actuarial chart?the portion that links games where the OBP was
.420-.429 and the SLG was also .420-.429.
It turns out that of the 130,000+ games in Tom’s database, less than 100 of
them match the criteria described above. (This isn’t surprising, actually, because
we’ve created very narrow ranges for both measures, and when you combine them,
it creates hundreds of mini-categories with relatively small sample sizes).
Just to give you a taste of that chart, here is a list of all the OBP/SLG combinations
with more than 500 games:
SLG?????????? OBP???? G? RUNS??? AVG
000-199?? 180-189? 1345?? 966?? 0.72
000-199?? 210-219? 1092?? 902?? 0.83
000-199?? 200-209?? 964?? 948?? 0.98
000-199?? 160-169?? 798?? 338?? 0.42
000-199?? 250-259?? 797? 1015?? 1.27
000-199?? 150-159?? 696?? 457?? 0.66
000-199?? 220-229?? 666?? 777?? 1.17
700+??????? ?500+?? 638? 8851? 13.87
000-199?? 230-239?? 608?? 667?? 1.10
000-199?? 130-139?? 558?? 197?? 0.35
000-199?? 120-129?? 548?? 289?? 0.53
As you can see, most of these games cluster at either the extreme low end (SLG
below .200 and OBP between .150-.250) or at the extreme high end (SLG .700 or
better, OBP .500 or better). As noted, the category breakdowns are extremely
At any rate, the 94 games in which a team produced an OBP and SLG that were
both in the .420-.429 range resulted in a total of 626 runs scored. You fundamentalists
out there may wish to take cover now, before I note that this works out to an
average of 6.66 runs per game.
Over the course of a 162-game season, that projects to 1079 runs.
That’s right about midway between the estimate using OBP times SLG plus 3.5%,
and the other estimates from those other theorists proclaiming that such a team
would score close to 1200 runs.
As Sean Forman pointed out, the OBP times SLG plus 3.5% formula actually works
pretty well if you remember to account for the extra plate appearances that
accrue when a team walks at a pace that is 30-40% higher than the current MLB
record for bases on balls in a season (835, by the 1949 Boston Red Sox).
However, a couple of other nuances remain. First is the fact that walks, despite
sabermetricians’ occasionally obsessive love for them, simply do not
have as much value in terms of run scoring as do hits. And we can demonstrate
this by examining the games within Tom Ruane’s data set that contain .420-.429
OBP and .420-.429 SLG.
I also asked Tom to segregate the OBP/SLG pairs by a third modifier?batting
average. How many runs per game did a team with less than a .310 BA score when
the OBP/SLG values in a game were between .420-.429?
How many when the team BA was .310-.349? And how many when the team BA was
.350 or higher?
While other analysts have noted that additional "secondary offensive"
characteristics have a strong tendency to enhance run scoring, that may not
be the case when dealing with more rarified levels of OBP. While two teams with
highly similar SLG but divergent batting averages quite often show a run scoring
advantage in favor of the team with the lower BA, it appears that the opposite
may be the case for OBP.
We can see this when we look at those 94 games in which teams produced an OBP/SLG
combination of .425/.425.? Less than one-fifth of these games (18, to be exact)
featured teams with BA’s lower than .310 (which would be in the range of the
team that Duane Thomas had selected, with its .290 BA). Breaking the run scoring
average for these games into groups based on BA, we see that the teams with
the lower BA scored fewer runs than was the case in the overall sample:
SLG?????????? OBP?????? BA?? G??? R?? R/G
420-429?? 420-429??? < 310? 18? 113? 6.28
420-429?? 420-429? 310-349? 50? 339? 6.78
420-429?? 420-429???? 350+? 26? 174? 6.69
420-429?? 420-429????? TOT? 94? 626? 6.66
Run scoring in games where OBP and SLG were highly similar (in the .400 to
.430 range) produced runs per game averages in the lower BA regions that averaged
somewhat more than four-tenths of a run lower than the average for these OBP/SLG
combinations as a whole.
There are two caveats here that need to be mentioned. First, the sample size
is small. Second, a team’s OBP/SLG stats are an aggregate of many games’ worth
of individual performances, and are not going to slavishly follow the results
in games that conform exactly to their overall performance level. (Mike Emeigh’s
research showed high OBP/low SLG teams exceeding their RC projection, but these
"high OBP" teams came from "aggregated" games as opposed
to those with an exact match in the .420-.429 OBP/SLG range. Mike’s high OBP/low
SLG sample produced an average of 6.82 runs per game.)
That said, it’s still an interesting effect, and it may go some distance toward
explaining why a walkman team with a .290/.425/.425 BA/OBP/SLG would tend to
score less runs than, say, a team with .330/.425/.425.
In short, the theorists who claim that BA is a superfluous statistic once you
have OBP and SLG would appear to have taken one too many liberties in their
zeal to create a calculational shorthand for run scoring. While for the most
part this shorthand works well, it appears to break down when we get to extremely
high OBP coupled with relatively low SLG (ie, SLG that is not as commensurately
high relative to the league as the team’s OBP).
This appears to be confirmed, at least provisionally, in the results for the
Tokyo Walkmen, a team based on Duane Thomas’ lineup selections that was
entered into competition at the sports simulation site WhatIf
Sports. After 144 games, the Walkmen?who were constrained
to play without a DH?have scored 814 runs, which puts them on pace to score
"only" 915 runs over 162 games.
Adding 10% to the Walkmen run total to adjust for the absence of the DH, we
see that our walk-taking pests would come in at approximately 1007 runs. That’s
about ten runs lower than what Tom Ruane’s empirical data suggests would be
the case for a .425/.425 OBP/SLG team with a sub-.300 BA (1017 runs).
And finally, there’s the issue raised earlier about just how much dropoff in
walk percentage (and, hence, in OBP) such a team would experience as a result
of having an entire lineup of walkmen. After some consideration, I managed to
come up with two benchmarks for this issue.
We’ll take the second one first, since we were talking about the What If Sports
simulation. While this is not even remotely an empirical test, looking at what
the simulation permitted the selected hitters to do is clearly a convenient
Interestingly, the simulation produced close-to-season-average OBP totals for
seven of the eight regulars who received signficant playing time (Wes Westrum,
catcher; Ferris Fain, first base; Eddie Stanky, second base; Eddie
Joost, shortstop; Eddie Yost, third base; Elmer Valo, left
field; Richie Ashburn, center field; Roy Cullenbine, right field).
One man?Elmer Valo?got hammered by the game. Valo’s .332 OBP (due in large
part to a .221 BA) brought the regulars’ OBP/SLG average down to .411/.410
(it was .419/.420 without him, which shows that What If Sports has a
pretty good simulator on its hands).
So the simulated team lost just under fifteen points of OPS. We can take a
more empirical approach, however. (And that brings us to the second of the two
nuances I referred to earlier.) Let’s look at one of the teams in baseball history
with the highest aggregate OBP?the 1921 Detroit Tigers. This team had
walkmen (Lu Blue, Donie Bush, Johnny Bassler) and high-average
hitters who walked at around league-average rates (the great outfield of Harry
Heilmann, Ty Cobb, and Bobby Veach).
The team OBP was .381; the eight regulars on the team (including second
baseman Ralph Young, who chipped in with the best season of his career,?
and third baseman Bob Jones, this team’s "weakest link") did
better than that, averaging an even .400.
What we want to know is?how much higher is the average of their peak OBP seasons?
When we look at Duane Thomas’ team, we see a lineup that features players performing
at their highest single-season level of OBP. "Mass career years" of
this nature are exceedingly rare in baseball (or in other endeavors as well).
It turns out that the best possible OBP for the eight regulars on the 1921
Tigers, using their best OBP seasons from their individual careers, works
out to .422.
So our data points indicate that this team lost somewhere between 14-22 points
of OBP to this "levelling off from career peak" effect. That’s about
a five percent dropoff.
Thus, for a team to have a realistic chance at producing a .425 OBP, it would
have to have players whose peak OBP performance would have to be that much higher?another
five percent above .425, or? close to .450 (.448 to be exact).
That’s if the Tiger example is representative, of course. I suspect it’s not
too far off, and it might actually be a little on the low side. An interesting
sabermetric study could be generated from this: calculate the percentage
of possible OBP achieved by teams, thus creating another benchmark for the
variability of season-to-season team performance. (You’d also want to do something
similar with SLG, of course.)
Prediction and projection tools have received a lot of exposure in the past
decade, but the basic "projection tool"?Runs Created and its many
competitors?is still for the most part ignoring the issue of season-to-season
variability. (The hoops that some analysts jump through to claim that they are
reducing the gap between projected and actual team run scoring have become downright
silly.) Getting deeper into the gradations of game-by-game performance is now
possible thanks to the increased access to detailed data, and it’s time for
theorists and those interested in such issues to embrace the possibilities that
are inherent in this approach, as opposed to re-inventing the wheel of "grand
theory" (what Bill James called the "great statistic"
in the 1984 Baseball Abstract).
I like ending on such a nice, bold thought?but then I remind myself that number-crunching
mania did not die off as a result of Bill’s words. If anything, the opposite
occured. As a matter of fact, Bill himself will be back with his own new and
improved "great statistic" later this year ("win shares").
And the beat goes on?
Posted: June 11, 2001 at 06:00 AM | 3 comment(s)
Login to Bookmark